Clinical deployment and prospective validation of an AI model for limb-length discrepancy measurements using an...

Area of Science:

• Artificial Intelligence in Medical Imaging
• Pediatric Orthopedics
• Radiology Workflow Optimization

Background:

• Limb-length discrepancy (LLD) is a common pediatric orthopedic condition requiring accurate measurement.
• Clinical deployment and prospective validation of AI models for LLD assessment are not well-established.
• Accurate LLD measurement is crucial for guiding treatment decisions and monitoring growth.

Purpose of the Study:

• To deploy an artificial intelligence (AI) model designed to measure limb-length discrepancy (LLD).
• To prospectively validate the performance of the deployed LLD AI model in a clinical setting.
• To assess the feasibility of integrating AI-driven LLD measurements into the radiology workflow.

Main Methods:

• An LLD AI model was containerized using Docker and integrated into a clinical computational platform.
• Two prospective validation studies were conducted: a shadow trial and a clinical trial using EOS scanograms.
• AI-generated bone lengths were compared against radiologist measurements using median absolute difference (MAD).

Main Results:

• The shadow trial (n=84) showed a MAD of 0.2 cm for both femoral and tibial lengths.
• The clinical trial (n=114) demonstrated a MAD of 0.3 cm for femoral and 0.2 cm for tibial lengths.
• The AI model achieved performance metrics comparable to those of radiologists in estimating bone lengths.

Conclusions:

• A computational platform enabled seamless deployment and integration of an LLD AI model into clinical workflow.
• Prospective validation confirmed the AI model's reliable performance in measuring limb-length discrepancy.
• The LLD AI model can function as a valuable secondary reader, enhancing diagnostic confidence and accuracy.